Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Multiple imputation showing missing values during imputation


From   "Nailing Xia" <nailingxia@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Multiple imputation showing missing values during imputation
Date   Thu, 11 Dec 2008 20:46:41 -0500

Hi,

I've done some experiment with the -interval- option in -ice- command.
And I found that the missing imputed value problem (i.e., Stata
indicating values for several cases of the variables imputed using
-intreg- are missing from previous cycle, see below email for more
information) goes away when I only include one interval variabel in
the interval option. The problem only occurs when there are 2 or more
variables included in the -interval- option. So I suspect that it
could be the syntax I wrote for the interval option is not correct. I
read the help menu for -ice-, but couldn't figure out what is wrong
with the syntax. So I am posting my code here--if you know the right
syntax, please help!

My code looks like (here I ommited some variables as well as options
including substitute, cmd, passive, saving, and weights):
ice age1 llage1 ulage1 age2 age3 llage3 ulage3 age4 age5 redlun4
llredlun4 ulredlun5 redlun5 llredlun5 ulredlun5.....interval(age1:
llage1 ulage1, age3: llage3 ulage3, redlun4: llredlun4 ulredlun5,
redlun5: llredlun5 ulredlun5)

Any thought or suggestions are very much appreciated!

Thanks,
Nailing Xia



On Tue, Dec 9, 2008 at 5:42 PM, Michael I. Lichter <mlichter@buffalo.edu> wrote:
> Nailing,
>
> Does -ice- provide information on each individual estimation? If so, you
> should check the -intreg- runs -- it may be that -intreg- is failing for
> some reason, possibly related to your interval() specifications, and that
> this is where the missing imputation data is coming from.
>
> Michael
>
> Nailing Xia wrote:
>>
>> Thanks, Michael. I checked the lower bound and upper bound variables
>> as well as the interval variables for cases that Stata report imputed
>> values being missing, but didn't find anything that looks strange. As
>> a matter of fact, those cases are within an interval (i.e., both the
>> lower bound and upper bound variables have a valid but different
>> value, indicating that the interval variable should be imputed within
>> that range) that needs an imputation with -intreg-.
>>
>> And here're the details of the dataset: I am using a longitudinal
>> dataset that follows over 10,000 children for several years. The
>> dataset has 5 waves, and in the model I included over 8,000
>> individuals. The -ice- command includes around 50 variables in
>> total--as some of the variables are time-specific but some are not,
>> the total number of variables are around 200. Options I used with
>> -ice- include "sub", "passive", "cmd", and "interval". For all imputed
>> variables (including categorical variables), I use -regress- (except
>> for those that are included in the "interval" option and thus use
>> -intreg-"). In "interval" option, I included 21 variables (out of over
>> 200 variables included in -ice-), and 15 of them have the imputed
>> value missing problem as described earlier. The number of missing
>> cases range from 1 to 49. What happens is that Stata will replace the
>> missing cases with mean and continue imputing. Then at cycle 6 when it
>> is imputing one of the variable that has the imputed value missing
>> problem, Stata shows that "convergence not achieved" and stops
>> imputing (although I am not sure whether this problem relates to the
>> imputed value missing problem).
>>
>> If you know anything about this problem, please help. Thanks a lot!
>>
>> Nailing Xia
>>
>>
>> On Tue, Dec 9, 2008 at 3:23 PM, Michael I. Lichter <mlichter@buffalo.edu>
>> wrote:
>>
>>>
>>> Nailing,
>>>
>>> I suspect that you haven't gotten any responses because your question is
>>> (or
>>> appears to be) very specific to your data, but you don't say much about
>>> the
>>> specifics of your data. It might be helpful to describe the variables
>>> involved and the constraints you are putting on the imputed data using
>>> the
>>> -interval- option to -ice-.
>>>
>>> Here's a stab in the dark, though. Is it possible that one or more of
>>> your
>>> interval variables are missing for two cases in which your variable
>>> redlun5
>>> is missing as well? If not, can you at least identify the two cases for
>>> which -ice- is failing to make an imputation and see if there is anything
>>> unusual about those cases' values on the other variables in the
>>> imputation?
>>>
>>> -ml
>>>
>>> Nailing Xia wrote:
>>>
>>>>
>>>> Hi,
>>>>
>>>> I am using -ice- for multiple imputation, and during the imputation
>>>> process, Stata shows that values for several imputed variables that
>>>> should have been imputed in previous cycle are actually missing. The
>>>> actual error message reads as: "I found that 2 missing value(s) of
>>>> redlun5 that should have been
>>>> imputed were missing at cycle 2. Replacing value(s) with mean, dumping
>>>> current data to file _ice_dump.dta and continuing."
>>>>
>>>> Anyone knows what seems to cause this problem? By the way, this
>>>> problem did not exist when I imputed without using the -interval-
>>>> option. But I checked my code for -interval- option, and it seems
>>>> fine.
>>>>
>>>> Many thanks!
>>>> Nailing Xia
>>>> *
>>>> *   For searches and help try:
>>>> *   http://www.stata.com/help.cgi?search
>>>> *   http://www.stata.com/support/statalist/faq
>>>> *   http://www.ats.ucla.edu/stat/stata/
>>>>
>>>>
>>>
>>> --
>>> Michael I. Lichter, Ph.D.
>>> Research Assistant Professor & NRSA Fellow
>>> UB Department of Family Medicine / Primary Care Research Institute
>>> UB Clinical Center, 462 Grider Street, Buffalo, NY 14215
>>> Office: CC 125 / Phone: 716-898-4751 / E-Mail: mlichter@buffalo.edu
>>>
>>> *
>>> *   For searches and help try:
>>> *   http://www.stata.com/help.cgi?search
>>> *   http://www.stata.com/support/statalist/faq
>>> *   http://www.ats.ucla.edu/stat/stata/
>>>
>>>
>
> --
> Michael I. Lichter, Ph.D.
> Research Assistant Professor & NRSA Fellow
> UB Department of Family Medicine / Primary Care Research Institute
> UB Clinical Center, 462 Grider Street, Buffalo, NY 14215
> Office: CC 125 / Phone: 716-898-4751 / E-Mail: mlichter@buffalo.edu
>
>
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index